基于BP神经网络的WSTi3515S阻燃钛合金超塑性变形行为预测

张学敏; 惠玉强; 李咪; 刘晓飞; 赖运金; 曹宇霞; 焦奔奇; 罗宇晴

doi:10.15980/j.tzzz.2019.06.023

轻有色合金 | 浏览量 : 0 下载量: 143 CSCD: 1

导出
分享
收藏
专辑

基于BP神经网络的WSTi3515S阻燃钛合金超塑性变形行为预测
Prediction of Superplastic Deformation Behavior of WSTi3515S Burn-resistant Titanium Alloy Based on BP Artificial Neural Network
2019年39卷第6期页码：668-671
纸质出版日期： 2019 ，
DOI： 10.15980/j.tzzz.2019.06.023

扫描看全文

[1]张学敏,惠玉强,李咪,刘晓飞,赖运金,曹宇霞,焦奔奇,罗宇晴.基于BP神经网络的WSTi3515S阻燃钛合金超塑性变形行为预测[J].特种铸造及有色合金,2019,39(06):668-671.

Zhang Xuemin, Xi Yuqiang, Li Mi, et al. Prediction of Superplastic Deformation Behavior of WSTi3515S Burn-resistant Titanium Alloy Based on BP Artificial Neural Network[J]. Special Casting & Nonferrous Alloys, 2019,39(6):668-671.
[1]张学敏,惠玉强,李咪,刘晓飞,赖运金,曹宇霞,焦奔奇,罗宇晴.基于BP神经网络的WSTi3515S阻燃钛合金超塑性变形行为预测[J].特种铸造及有色合金,2019,39(06):668-671. DOI： 10.15980/j.tzzz.2019.06.023.

Zhang Xuemin, Xi Yuqiang, Li Mi, et al. Prediction of Superplastic Deformation Behavior of WSTi3515S Burn-resistant Titanium Alloy Based on BP Artificial Neural Network[J]. Special Casting & Nonferrous Alloys, 2019,39(6):668-671. DOI： 10.15980/j.tzzz.2019.06.023.

摘要

以WSTi3515S阻燃钛合金超塑性拉伸试验所得试验数据为基础

基于BP （Back Propagation）人工神经网络方法

建立了该合金超塑性变形过程中流变应力的神经网络预测模型。模型的输入参数为变形温度、应变和应变速率;输出参数为流变应力;结构为3×10×12×1双隐含层的4层网络模型。结果表明

BP神经网络模型能够很好地描述WSTi3515S合金超塑性变形时各热力学参数之间高度非线性的复杂关系

可精确地预测超塑性拉伸过程中的流变应力

平均相对误差为1.08%。

Abstract

Based on the experimental data of superplastic tensile tests

the neural network prediction model of flow stress of WSTi3515 Sburn-resistant titanium alloy during superplastic deformation was established using BP artificial neural network.The inputs of the model include deformation temperature

strain and strain rate

while the output of the model is the flow stress.The structure adopted four layers back propagation neural network model with 3×10×12×1 double hidden layers.The results show that the BP neural network model can well describe the highly nonlinear complex relationship among the thermodynamic parameters of WSTi3515 Salloy during superplastic deformation.It also can accurately predict the flow stress during superplastic tensile tests

and the average relative error is 1.08%.